TY - JOUR
T1 - A novel approach of collision assessment for coastal radar surveillance
AU - Ma, Feng
AU - Chen, Yu-Wang
AU - Huang, Zi-chao
AU - Yan, Xin-ping
AU - Wang, Jin
PY - 2016/11
Y1 - 2016/11
N2 - For coastal radar surveillance, this paper proposes a data-driven approach to estimate a blip's collision probability preliminarily based on two factors: the probability of it being a moving vessel and the collision potential of its position. The first factor is determined by a Directed Acyclic Graph (DAG), whose nodes represent the blip's characteristics, including the velocity, direction and size. Additionally, the structure and conditional probability tables of the DAG can be learned from verified samples. Subsequently, obstacles in a waterway can be described as collision potential fields using an Artificial Potential Field model, and the corresponding coefficients can be trained in accordance with the historical vessel distribution. Then, the other factor, the positional collision potential of any position is obtained through overlapping all the collision potential fields. For simplicity, only static obstacles have been considered. Eventually, the two factors are characterised as evidence, and the collision probability of a blip is estimated by combining them with Dempster's rule. Through ranking blips on collision probabilities, those that pose high threat to safety can be picked up in advance to remind radar operators. Particularly, a good agreement between the proposed approach and the manual operation was found in a preliminary test.
AB - For coastal radar surveillance, this paper proposes a data-driven approach to estimate a blip's collision probability preliminarily based on two factors: the probability of it being a moving vessel and the collision potential of its position. The first factor is determined by a Directed Acyclic Graph (DAG), whose nodes represent the blip's characteristics, including the velocity, direction and size. Additionally, the structure and conditional probability tables of the DAG can be learned from verified samples. Subsequently, obstacles in a waterway can be described as collision potential fields using an Artificial Potential Field model, and the corresponding coefficients can be trained in accordance with the historical vessel distribution. Then, the other factor, the positional collision potential of any position is obtained through overlapping all the collision potential fields. For simplicity, only static obstacles have been considered. Eventually, the two factors are characterised as evidence, and the collision probability of a blip is estimated by combining them with Dempster's rule. Through ranking blips on collision probabilities, those that pose high threat to safety can be picked up in advance to remind radar operators. Particularly, a good agreement between the proposed approach and the manual operation was found in a preliminary test.
U2 - 10.1016/j.ress.2016.07.013
DO - 10.1016/j.ress.2016.07.013
M3 - Article
SN - 0951-8320
VL - 155
SP - 179
EP - 195
JO - Reliability Engineering and System Safety
JF - Reliability Engineering and System Safety
ER -